The
cardiovascular PhysioLab platform and its applications in toxicity
prediction
Héctor de Léon, Entelos
Entelos has developed a set of biosimulation and gene expression
profiling tools aimed at the early identification of effective drug
candidates with low toxicity. Its main strength is in a dynamic
representation of whole-body lipoprotein synthesis, distribution,
processing and uptake. The company has assembled the cardiovascular
PhysioLab platform, a large-scale mathematical model of human lipid
metabolism and cardiovascular pathology, to evaluate the potential
efficacy of alternative therapeutic approaches. The model uses
differential equations to represent interactions of cells and
biomolecules linked to key cardiovascular clinical outcomes such as
myocardial infarction. Finite-element modeling is used to simulate the
temporal changes in the structure of atherosclerotic plaques that lead
to rupture. The structural stability of the plaque can be linked to an
estimated risk of a cardiovascular event. Virtual patients and patient
populations are used to represent different pathophysiological
hypotheses and to analyze the impact of phenotypic variability in
response to therapies and drug-induced toxicity. The PhysioLab platform
has been validated against data from a number of clinical trials.
De Léon presented a case study of identification of novel
candidate biomarkers for patient stratification. Raising high-density
lipoprotein cholesterol (HDL-C) is a promising strategy in prevention
of cardiovascular disease, and cholesteryl ester transfer protein
(CETP) inhibitors have been developed to reduce atherosclerosis.
Entelos used a biologically diverse cohort of 60 virtual patients and
simulated their response, after two years, to treatment with either a
statin or a statin plus a CETP inhibitor. Only a third of patients
responded to CETP inhibition, determined by percent atheroma volume.
Baseline lipids did not correlate with response to CETP inhibitor
treatment. A novel candidate multivariate biomarker was identified for
exclusion of CETP adverse responders prior to treatment.
Entelos can identify novel, optimal collections of measurements
(candidate multivariate biomarkers) predictive of efficacy and/or
safety; determine different classes of biomarkers; provide assessments
of biomarker robustness (using sensitivity and specificity, and R2) and
optimality; and provide recommendations for means to validate candidate
biomarkers.
The company has also developed DrugMatrix, a toxicogenomic database of
microarray expression data linked to classic preclinical and clinical
toxicology measurements, to identify predictive gene expression
profiles. Hypotheses generated from these profiles can be simulated in
the cardiovascular PhysioLab platform to identify biomarkers predictive
of adverse events. De Léon presented a case study
identifying
putative mechanisms to differentiate efficacy and safety of two
compounds.
Peroxisome proliferator-activated receptor (PPAR) agonists, such as
Actos (pioglitazone) and Avandia (rosiglitazone), activate nuclear
hormone receptors which improves insulin sensitivity. De
Léon
hypothesized that reported differences in lipoprotein particle
distributions in patients treated with Actos versus Avandia correspond
with associated differences in hepatic gene expression. He queried the
DrugMatrix database and compared hepatic gene expression profiles from
animals treated with Actos or Avandia, and established differences
between Actos- and Avandia-induced changes in gene expression. When
administered in high doses sub-chronically, Actos and Avandia evoke
differential patterns in hepatic gene expression. Entelos is still
analyzing the significance of these results. De Léon also
hypothesized that the reported differences in lipoprotein particle
distributions in patients treated with Actos versus Avandia yield
differential effects on plaque growth and stability, and he presented
evidence for his hypothesis.
In the discussion session, the chairman was concerned about
incorporation of exposure and dose rate; an attendee from GSK commented
that this was not really the forum for making unsubstantiated (or
non-validated) claims in public about a marketed drug [Avandia]; and
another attendee asked whether we need to understand differential
equations (asking about the finite element modeling system used by the
PhysioLab platform to simulate plaque rupture).
Emerging
areas in toxicity prediction: an NIHS perspective
Akihiko Hirose, National Institute of Health Sciences (NIHS), Japan
We urgently need to develop a high throughput evaluation system for the
risk to humans of environmental chemicals. Since no individual QSAR
system is powerful enough, NIHS has started to develop a workflow to
assess genotoxicity using a combination of three in silico
systems: Derek for Windows (a rule-based system), MCASE (a database and
substructure-based system) and ADMEWorks (an unsupervised regression
classification system from Fujitsu Kyushu System Engineering).31
Each system was customized for mutagenicity prediction, using bacterial
gene mutation and in
vitro chromosomal aberration assays. In a combination
approach, the concordance between in
vitro and in
silico
assays on bacterial gene mutation reached around 94%, although
applicability decreased to 55%. Next, NIHS tried a similar approach for
developing a chromosomal aberration prediction system. The performance
in this case was even lower than that of bacterial mutagenicity
prediction and further development is required.
In addition to these genotoxicity studies, repeated dose rat toxicity
studies are commonly used to evaluate the risk of industrial chemicals,
but no suitable in
silico
general toxicity evaluation system is available at present. NIHS
analyzed the toxicity profiles of hundreds of 28-day repeated dose
studies and focused on developing a prediction system for
hepatotoxicity and/or renal toxicity endpoints, by searching new
substructural alerts for Derek for Windows and using Leadscope
Predictive Data Miner, a discriminant-based QSAR model builder. Rapid
alerts are being developed for Derek for Windows in order to improve
the sensitivity, although this may cause an increase in the number of
false positives. With Leadscope Predictive Data Miner high concordance
models could be obtained by using a consensus approach or by
restricting the probability thresholds, although the applicability was
decreased to about 40-50%. With the ADMEWorks model builder a high
concordance model to predict liver weight changes, using an SVM method,
was obtained as a single prediction model but other models had
relatively low concordance. In order to improve predictability, a
combination approach of Derek and the statistical data mining models
would be required. In addition, more accurate structural alerts and
endpoint-specific prediction models could be constructed by using a
more precise learning data set.
NIHS has also joined a multi-institutional Japanese project developing
a repeated-dose toxicity knowledge base system, which could assist
toxicological expert judgment, or support preliminary governmental
decisions. The system consists of three parts: a detailed subchronic
toxicity studies database, a toxicity mechanisms database, and a
metabolite prediction system. The project is led by the National
Institute of Technology and Evaluation (NITE), Tohoku University,
Kwansei Gakuin University, Fujitsu Co. Ltd., and NIHS. Parts of this in silico
knowledge based system will be integrated in the OECD (Q)SAR
Application Tool Box,7 and will support a categorical approach to
evaluation of high production volume chemicals. A repeated-dose
toxicity (Q)SAR system will also be developed in future.
